A gradient of childhood self-control predicts health, wealth, and public safety

When the children reached the age of 32 y, we assessed their cardiovascular, respiratory, dental, and sexual health as well as their inflammatory status by carrying out physical examinations and laboratory tests to assess metabolic abnormalities (including overweight), airflow limitation, periodontal disease, sexually transmitted infection, and C-reactive protein level, respectively. We summed these five clinical measures into a simple physical health index for each study member: 43% of study members had none of the biomarkers, 37% had one, and 20% had two or more. Childhood self-control predicted adult health problems (Table 1, model 1), even after accounting for social class origins and IQ (Table 1, model 2). SI Appendix, Table S1 shows associations between self-control and each individual health measure.

We also conducted clinical interviews with the study members at the age of 32 y to assess depression and substance dependence (tobacco, alcohol, and cannabis dependence as well as dependence on other street and prescription drugs), following the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV) criteria (24). As adults, children with poor self-control were not at elevated risk for depression. They had elevated risk for substance dependence (Table 1, model 1), however, even after accounting for social class and IQ (Table 1, model 2). This longitudinal link between self-control and substance dependence was verified by people whom study members had nominated as informants who knew them well. As adults, children with poor self-control were rated by their informants as having alcohol and drug problems (Table 1).

Childhood self-control also foreshadowed the study members’ financial situations. Although the study members’ social class of origin and IQ were strong predictors of their adult socioeconomic status and income, poor self-control offered significant incremental validity in predicting the socioeconomic position they achieved and the income they earned (Table 1). By the age of 32 y, 47% of study members had become parents. Childhood self-control predicted whether or not these study members’ offspring were being reared in one-parent vs. two-parent households (e.g., the study member was an absent father or single mother), also after accounting for social class and IQ (Table 1).

At the age of 32 y, children with poor self-control were less financially planful. Compared with other 32-y-olds, they were less likely to save and had acquired fewer financial building blocks for the future (e.g., home ownership, investment funds, retirement plans). Children with poor self-control were also struggling financially in adulthood. They reported more money-management difficulties and had accumulated more credit problems (Table 1). Poor self-control in childhood was a stronger predictor of these financial difficulties than study members’ social class origins and IQ. This longitudinal link between self-control and self-reported financial problems was verified by informants who knew them well. As adults, children with poor self-control were rated by their informants as poor money managers (Table 1).

We obtained records of study members’ court convictions at all courts in New Zealand and Australia by searching the central computer systems of the New Zealand Police; 24% of the study members had been convicted of a crime by the age of 32 y. Children with poor self-control were more likely to be convicted of a criminal offense, even after accounting for social class origins and IQ (Table 1).

We observed a self-control gradient in which boys and girls with less self-control had worse health, less wealth, and more crime as adults than those with more self-control at every level of the distribution of self-control (Fig. 2). To document further that self-control relates to outcomes all along its gradient, we removed 61 study members who were diagnosed with attention-deficit hyperactivity disorder (a childhood psychiatric disorder of impaired impulse control) and repeated this analysis. The gradient associations in Fig. 2 remained unaltered. In addition, we tested whether self-control effects operate throughout the distribution or are confined to the least self-controlled children. We repeated analyses after removing children in the least self-controlled quintile and continued to observe significant linear associations. The self-control gradient was even apparent when we removed children in the least and most self-controlled quintiles (SI Appendix, Table S2).

An interesting question is what would happen if we were able to intervene and improve children's self-control. Would an increase in self-control predict better outcomes? Although the study did not include an experimental intervention, we were able to address this question by studying children who moved up the rank in their self-control from childhood to young adulthood. The childhood measure of self-control was significantly correlated with a personality measurement of self-control administered to our cohort in young adulthood (r = 0.30, P ≤ 0.001), at a moderate magnitude, consistent with expectations (25) (SI Appendix). This stability coefficient implies that some children also changed their rank order in self-control. Moreover, those children who became more self-controlled from childhood to young adulthood had better outcomes by the age of 32 y, even after controlling for their initial levels of childhood self-control (SI Appendix, Table S3). As a caveat, it is not clear that natural history change of the sort we observed in our longitudinal study is equivalent to intervention-induced change. Nevertheless, these suggestive findings should stimulate consideration of interventions to raise self-control.

Data collected at the ages of 13, 15, 18, and 21 y showed that children with poor self-control were more likely to make mistakes as adolescents, resulting in “snares” that trapped them in harmful lifestyles. More children with low self-control began smoking by the age of 15 y [odds ratio (OR) = 1.69, 95% confidence interval (CI): 1.45–1.96], left school early with no educational qualifications (OR = 2.28, 95% CI: 1.92–2.72), and became unplanned teenaged parents (OR = 1.79, 95% CI: 1.40–2.29). The lower their self-control, the more of these snares they encountered (incident rate ratio = 1.48, 95% CI: 1.38–1.59) (SI Appendix, Fig. S1). In turn, the more snares they encountered, the more likely they were, as adults, to have poor health, less wealth, and criminal conviction (SI Appendix, Table S4). We tested whether snares explained the long-term predictive power of self-control in two ways. First, using statistical controls, we partialled out the portion of the association between childhood self-control and each adult outcome that was accounted for by adolescent snares. The snares attenuated the effect of self-control on health by 32%, substance dependence by 63%, socioeconomic status by 35%, income by 29%, single-parent child rearing by 61%, financial planfulness by 35%, financial struggles by 47%, and crime by 42%. The direct effect of self-control remained statistically significant for nearly every outcome measure, however (SI Appendix, Table S4). Second, we tested the association between childhood self-control and the adult outcomes among adolescents who did not encounter any snares, a so-called “utopian” control group (26). Again, prediction from childhood self-control to the adult measures remained significant even among this group of nonsmoking, non–teen-parent, high-school graduates (SI Appendix, Table S4).

Our composite measure of self-control in the Dunedin study included assessments from the age of 3–11 y. To answer this question, we isolated staff ratings of the children's self-control observed during 90-min data collection sessions at the research unit in the mid-1970s, when they were 3–5 y old (27). This standardized observational measure of preschoolers’ self-control significantly predicted health, wealth, and convictions at the age of 32 y, albeit with modest effect sizes (SI Appendix, Table S5).

In the Dunedin study, statistical controls revealed that self-control had its own associations with outcomes, apart from childhood social class and IQ. Each Dunedin study member grew up in a different family, however, and their families varied widely on many features that affect children's outcomes. A compelling quasiexperimental research design that can isolate the influence of self-control is to track and compare siblings. Does the sibling with poorer self-control have worse outcomes than his or her more self-controlled sibling growing up in the same family environment? To apply this design, we turned to a second sample, the Environmental-Risk Longitudinal Twin Study (E-Risk), where we have been tracking a birth cohort of British twins since their birth in 1994 to 1995 with 96% retention (SI Appendix). When the E-Risk study twins were 5 y old, research staff rated each child on the same observational measure of self-control originally used with Dunedin study children as preschoolers. Although the E-Risk study children have been followed only up to age of 12 y, their self-control already forecasts many of the adult outcomes we saw in the Dunedin study. We applied sibling fixed-effects models to same-gender dizygotic pairs (n = 509 pairs) because they are no more alike than ordinary siblings (with the added advantages of being the same age and gender). Models showed that the 5-y-old sibling with poorer self-control was significantly more likely to begin smoking as a 12-y-old (a precursor of adult ill health; B = 0.07, SE = 0.003; P < 0.03), perform poorly in school (a precursor of adult wealth; B = −0.13, SE = 0.007; P < 0.001), and engage in antisocial behaviors (a precursor of adult crime; B = 0.09, SE = 0.007; P = 0.007), and these findings remained significant even after controlling for sibling differences in IQ (B = 0.07, SE = 0.003, P = 0.02 for smoking; B = −0.07, SE = 0.006, P = 0.01 for school performance; and B = 0.09, SE = 0.007, P = 0.005 for antisocial behavior).

Participants are members of the Dunedin Multidisciplinary Health and Development Study, which tracks the development of 1,037 individuals born in 1972–1973 in Dunedin, New Zealand.

Children's self-control during their first decade of life was measured using nine measures of self-control: observational ratings of children's lack of control (3 and 5 y of age) and parent, teacher, and self-reports of impulsive aggression, hyperactivity, lack of persistence, inattention, and impulsivity (5, 7, 9, and 11 y of age). The nine measures were positively and significantly correlated. Based on principal components analysis, the standardized measures were averaged into a single composite score (M = 0, SD = 1), comprising multiple ages and informants, with strong internal reliability α = 0.86. SI Appendix, Table S6 shows that whether we examined self-control as measured by observers, teachers, parents, or children's self-reports, individual differences in childhood self-control were significantly related to each of the adult health, wealth, and public safety outcomes; that is, the results were not sensitive to the use of any particular source of information about children's self-control and were robust to the data source in measuring self-control.

Health, wealth, and crime outcomes were assessed at age 32 y by physical examinations, blood tests, personal interviews, record searches, and informant reports.

Participants are members of the E-Risk study, which tracks the development of a nationally representative birth cohort of 2,232 twin children born in England and Wales in 1994–1995.

After completing the home visit when siblings were 5 y of age, examiners rated each twin on the measure of self-control that was originally used in the Dunedin study when the children in that study were 3 and 5 y of age (27). In this assessment procedure, the examiners evaluated the following behaviors: lability, low frustration tolerance, hostility, roughness, resistance, restlessness, impulsivity, fleeting attention, and lacking persistence. Each behavioral characteristic was defined in explicit terms, and the examiner evaluated whether each characteristic was not at all (0), somewhat (1), or definitely characteristic (2) of the child. The (interrater) reliability was 0.79.

Children reported about their delinquent behavior and smoking. Children's educational performance was evaluated by their teachers, who rated each child's performance in English and mathematics.